haojae

Three tasks: automate discovery, verification, and insertion of academic citations into LaTeX and Word manuscripts; transform scientific drafts into submission-grade English or Chinese prose; and produce publication-grade data visualizations with chart-type recommendation, error interception, and bilingual rendering.

Researchers preparing journal figures who are unsure which chart type fits their data. The skill profiles columns, sample sizes, distributions, outliers, and correlations, recommends appropriate chart types, blocks classic mistakes like pie charts and dual Y-axes, and outputs Nature, Science, IEEE, Elsevier, or PNAS-grade figures.

It covers line, bar, scatter, box/violin, heatmap, error-bar, histogram/KDE distribution, correlation matrix, scatter matrix, and multi-panel figures. The stack is matplotlib plus seaborn plus SciencePlots for static output and plotly for interactive charts, with vector export and significance annotation.

Yes. It is fully bilingual and in Chinese mode auto-configures Noto Sans CJK, Source Han Sans, or SimHei fonts, fixes the minus-sign box glyph issue, and supports the mixed typesetting of SimSun body text with Times New Roman numerals required by Chinese core journals.

It runs a visual self-check loop: renders a PNG preview, programmatically detects missing glyphs, text clipping, and tick overlap, then performs an image review for occlusion and subplot alignment, re-rendering until all checks pass. Default output uses colorblind-safe palettes with redundant encoding and grayscale preview.